v1.1.6

v1.1.6

Context Strategies!

Three options to control how much the server seeds into context:

• full: everything (~42k tokens)
• lean: action names visible, descriptions on demand (~23k tokens)
• micro: one gateway (~1k tokens)

full remains the default selection for now, but this will change as I refine the behavior of the smaller contexts.

Switch between them with ue-mcp context [ full | lean | micro ] or in the ue-mcp init flow's Context Strategy page.

Server

• micro strategy. Collapses the entire surface behind a single tools gateway (list_categories, describe, call), mirroring the native MCP toolset gateway. Category tools stay unadvertised; call(category, method, args) dispatches straight to the target action. Seed drops from ~45k tokens (full) to ~1k on the test project. Set it with ue-mcp context micro, the init Context strategy page, context.strategy: micro in ue-mcp.yml, or UE_MCP_CONTEXT_STRATEGY=micro.
• Slimmer full. Server instructions no longer re-list every category's actions, since that already lives in each tool's description and schema. Saves ~1.4k tokens from the default tier with no capability loss.

Internals

• npm run context-tax benchmarks all three tiers in one run, reporting each tier's token count and its reduction versus full. Counts are exact via the Anthropic count_tokens endpoint when ANTHROPIC_API_KEY is set.
• The task registry keeps every category task in all modes, so flows resolve regardless of the advertised surface.